Improved Structural Discovery and Representation Learning of Multi-Agent Data
This work addresses representation learning challenges in multi-agent systems, such as sports analytics, but appears incremental as it builds on existing methods for structured data.
The paper tackles the problem of representing multi-agent data with latent group structures by introducing a dynamic alignment method that learns group structure and orders agents consistently, enabling faster representation learning, as demonstrated on professional soccer tracking data with unspecified speed improvements.
Central to all machine learning algorithms is data representation. For multi-agent systems, selecting a representation which adequately captures the interactions among agents is challenging due to the latent group structure which tends to vary depending on context. However, in multi-agent systems with strong group structure, we can simultaneously learn this structure and map a set of agents to a consistently ordered representation for further learning. In this paper, we present a dynamic alignment method which provides a robust ordering of structured multi-agent data enabling representation learning to occur in a fraction of the time of previous methods. We demonstrate the value of this approach using a large amount of soccer tracking data from a professional league.